Comparison of EfficientNet and VGGNet Using RGB Photos for Human Action Recognition

In the field of computer vision, human action recognition, is a key technology, underpinning numerous application scenarios such as video surveillance, human-computer interaction, sports analysis, and more. Convolutional Neural Network (CNN)-based models have demonstrated impressive performance in human action detection difficulties due to the swift advancement of deep learning techniques. The subject matter focuses on the research of human action identification using two advanced CNN models, VGGNet and EfficientNet, on RGB image datasets. It also compares the differences in feature extraction and classification capabilities between the VGGNet and EfficientNet models. This paper selects an RGB image dataset containing information on 15 categories of human actions as the experimental basis, providing strong support for model training. During the evaluation phase, accuracy is adopted as the metric, and each model's capabilities undergo evaluation. The results indicate that the EfficientNet model exhibits higher accuracy and better generalization capabilities in human action recognition tasks, primarily due to its efficient network structure and optimized model design. The categorization of various postures aids in the advancement of technology for recognizing human behavior and helps to highlight the diversity and complexity of human behavior.